This is an archive of the discontinued LLVM Phabricator instance.

Differential D16297

[AVX512] Adding VPERMT2B and VPERMI2B instruction .
ClosedPublic

Authored by m_zuckerman on Jan 18 2016, 9:02 AM.

Details

Reviewers
AsafBadouh
delena
igorb
Commits
rG4582bdab12c0: [AVX512] Adding VPERMT2B and VPERMI2B instruction .
rL258161: [AVX512] Adding VPERMT2B and VPERMI2B instruction .

Diff Detail

Event Timeline

m_zuckerman updated this revision to Diff 45191.Jan 18 2016, 9:02 AM
m_zuckerman retitled this revision from to D16294: [AVX512] Adding VPERMT2B and VPERMI2B instruction ..
m_zuckerman updated this object.
m_zuckerman added reviewers: AsafBadouh, delena, igorb.
m_zuckerman added a subscriber: llvm-commits.
m_zuckerman updated this revision to Diff 45194.Jan 18 2016, 9:25 AM
delena accepted this revision.Jan 18 2016, 11:51 AM
delena edited edge metadata.

LGTM

This revision is now accepted and ready to land.Jan 18 2016, 11:51 AM
m_zuckerman retitled this revision from D16294: [AVX512] Adding VPERMT2B and VPERMI2B instruction . to [AVX512] Adding VPERMT2B and VPERMI2B instruction ..Jan 19 2016, 10:49 AM
m_zuckerman edited edge metadata.
Closed by commit rL258161: [AVX512] Adding VPERMT2B and VPERMI2B instruction . (authored by mzuckerm). · Explain WhyJan 19 2016, 10:51 AM
This revision was automatically updated to reflect the committed changes.

Revision Contents

PathSize
lib/
Target/
X86/
34 lines
test/
MC/
X86/
145 lines

Diff 45191

lib/Target/X86/X86InstrAVX512.td

  • This file is larger than 256 KB, so syntax highlighting is disabled by default.
Show First 20 Lines • Show All 1,172 Lines • ▼ Show 20 Lines defm NAME#128: avx512_perm_i<opc, OpcodeStr, VTInfo.info128,
ShuffleMask.info128>, EVEX_V128; ShuffleMask.info128>, EVEX_V128;
defm NAME#256: avx512_perm_i<opc, OpcodeStr, VTInfo.info256, defm NAME#256: avx512_perm_i<opc, OpcodeStr, VTInfo.info256,
ShuffleMask.info256>, ShuffleMask.info256>,
avx512_perm_i_mb<opc, OpcodeStr, VTInfo.info256, avx512_perm_i_mb<opc, OpcodeStr, VTInfo.info256,
ShuffleMask.info256>, EVEX_V256; ShuffleMask.info256>, EVEX_V256;
} }
} }
multiclass avx512_perm_i_sizes_w<bits<8> opc, string OpcodeStr, multiclass avx512_perm_i_sizes_bw<bits<8> opc, string OpcodeStr,
AVX512VLVectorVTInfo VTInfo, AVX512VLVectorVTInfo VTInfo,
AVX512VLVectorVTInfo Idx> { AVX512VLVectorVTInfo Idx,
let Predicates = [HasBWI] in Predicate Prd> {
let Predicates = [Prd] in
defm NAME: avx512_perm_i<opc, OpcodeStr, VTInfo.info512, defm NAME: avx512_perm_i<opc, OpcodeStr, VTInfo.info512,
Idx.info512>, EVEX_V512; Idx.info512>, EVEX_V512;
let Predicates = [HasBWI, HasVLX] in { let Predicates = [Prd, HasVLX] in {
defm NAME#128: avx512_perm_i<opc, OpcodeStr, VTInfo.info128, defm NAME#128: avx512_perm_i<opc, OpcodeStr, VTInfo.info128,
Idx.info128>, EVEX_V128; Idx.info128>, EVEX_V128;
defm NAME#256: avx512_perm_i<opc, OpcodeStr, VTInfo.info256, defm NAME#256: avx512_perm_i<opc, OpcodeStr, VTInfo.info256,
Idx.info256>, EVEX_V256; Idx.info256>, EVEX_V256;
} }
} }
defm VPERMI2D : avx512_perm_i_sizes<0x76, "vpermi2d", defm VPERMI2D : avx512_perm_i_sizes<0x76, "vpermi2d",
avx512vl_i32_info, avx512vl_i32_info>, EVEX_CD8<32, CD8VF>; avx512vl_i32_info, avx512vl_i32_info>, EVEX_CD8<32, CD8VF>;
defm VPERMI2Q : avx512_perm_i_sizes<0x76, "vpermi2q", defm VPERMI2Q : avx512_perm_i_sizes<0x76, "vpermi2q",
avx512vl_i64_info, avx512vl_i64_info>, VEX_W, EVEX_CD8<64, CD8VF>; avx512vl_i64_info, avx512vl_i64_info>, VEX_W, EVEX_CD8<64, CD8VF>;
defm VPERMI2W : avx512_perm_i_sizes_w<0x75, "vpermi2w", defm VPERMI2W : avx512_perm_i_sizes_bw<0x75, "vpermi2w",
avx512vl_i16_info, avx512vl_i16_info>, VEX_W, EVEX_CD8<16, CD8VF>; avx512vl_i16_info, avx512vl_i16_info, HasBWI>, VEX_W,
EVEX_CD8<16, CD8VF>;
defm VPERMI2B : avx512_perm_i_sizes_bw<0x75, "vpermi2b",
avx512vl_i8_info, avx512vl_i8_info, HasVBMI>,
EVEX_CD8<8, CD8VF>;
defm VPERMI2PS : avx512_perm_i_sizes<0x77, "vpermi2ps", defm VPERMI2PS : avx512_perm_i_sizes<0x77, "vpermi2ps",
avx512vl_f32_info, avx512vl_i32_info>, EVEX_CD8<32, CD8VF>; avx512vl_f32_info, avx512vl_i32_info>, EVEX_CD8<32, CD8VF>;
defm VPERMI2PD : avx512_perm_i_sizes<0x77, "vpermi2pd", defm VPERMI2PD : avx512_perm_i_sizes<0x77, "vpermi2pd",
avx512vl_f64_info, avx512vl_i64_info>, VEX_W, EVEX_CD8<64, CD8VF>; avx512vl_f64_info, avx512vl_i64_info>, VEX_W, EVEX_CD8<64, CD8VF>;
// VPERMT2 // VPERMT2
multiclass avx512_perm_t<bits<8> opc, string OpcodeStr, multiclass avx512_perm_t<bits<8> opc, string OpcodeStr,
X86VectorVTInfo _, X86VectorVTInfo IdxVT> { X86VectorVTInfo _, X86VectorVTInfo IdxVT> {
Show All 39 Lines defm NAME#128: avx512_perm_t<opc, OpcodeStr, VTInfo.info128,
ShuffleMask.info128>, EVEX_V128; ShuffleMask.info128>, EVEX_V128;
defm NAME#256: avx512_perm_t<opc, OpcodeStr, VTInfo.info256, defm NAME#256: avx512_perm_t<opc, OpcodeStr, VTInfo.info256,
ShuffleMask.info256>, ShuffleMask.info256>,
avx512_perm_t_mb<opc, OpcodeStr, VTInfo.info256, avx512_perm_t_mb<opc, OpcodeStr, VTInfo.info256,
ShuffleMask.info256>, EVEX_V256; ShuffleMask.info256>, EVEX_V256;
} }
} }
multiclass avx512_perm_t_sizes_w<bits<8> opc, string OpcodeStr, multiclass avx512_perm_t_sizes_bw<bits<8> opc, string OpcodeStr,
AVX512VLVectorVTInfo VTInfo, AVX512VLVectorVTInfo VTInfo,
AVX512VLVectorVTInfo Idx> { AVX512VLVectorVTInfo Idx,
let Predicates = [HasBWI] in Predicate Prd> {
let Predicates = [Prd] in
defm NAME: avx512_perm_t<opc, OpcodeStr, VTInfo.info512, defm NAME: avx512_perm_t<opc, OpcodeStr, VTInfo.info512,
Idx.info512>, EVEX_V512; Idx.info512>, EVEX_V512;
let Predicates = [HasBWI, HasVLX] in { let Predicates = [Prd, HasVLX] in {
defm NAME#128: avx512_perm_t<opc, OpcodeStr, VTInfo.info128, defm NAME#128: avx512_perm_t<opc, OpcodeStr, VTInfo.info128,
Idx.info128>, EVEX_V128; Idx.info128>, EVEX_V128;
defm NAME#256: avx512_perm_t<opc, OpcodeStr, VTInfo.info256, defm NAME#256: avx512_perm_t<opc, OpcodeStr, VTInfo.info256,
Idx.info256>, EVEX_V256; Idx.info256>, EVEX_V256;
} }
} }
defm VPERMT2D : avx512_perm_t_sizes<0x7E, "vpermt2d", defm VPERMT2D : avx512_perm_t_sizes<0x7E, "vpermt2d",
avx512vl_i32_info, avx512vl_i32_info>, EVEX_CD8<32, CD8VF>; avx512vl_i32_info, avx512vl_i32_info>, EVEX_CD8<32, CD8VF>;
defm VPERMT2Q : avx512_perm_t_sizes<0x7E, "vpermt2q", defm VPERMT2Q : avx512_perm_t_sizes<0x7E, "vpermt2q",
avx512vl_i64_info, avx512vl_i64_info>, VEX_W, EVEX_CD8<64, CD8VF>; avx512vl_i64_info, avx512vl_i64_info>, VEX_W, EVEX_CD8<64, CD8VF>;
defm VPERMT2W : avx512_perm_t_sizes_w<0x7D, "vpermt2w", defm VPERMT2W : avx512_perm_t_sizes_bw<0x7D, "vpermt2w",
avx512vl_i16_info, avx512vl_i16_info>, VEX_W, EVEX_CD8<16, CD8VF>; avx512vl_i16_info, avx512vl_i16_info, HasBWI>,
VEX_W, EVEX_CD8<16, CD8VF>;
defm VPERMT2B : avx512_perm_t_sizes_bw<0x7D, "vpermt2b",
avx512vl_i8_info, avx512vl_i8_info, HasVBMI>,
EVEX_CD8<8, CD8VF>;
defm VPERMT2PS : avx512_perm_t_sizes<0x7F, "vpermt2ps", defm VPERMT2PS : avx512_perm_t_sizes<0x7F, "vpermt2ps",
avx512vl_f32_info, avx512vl_i32_info>, EVEX_CD8<32, CD8VF>; avx512vl_f32_info, avx512vl_i32_info>, EVEX_CD8<32, CD8VF>;
defm VPERMT2PD : avx512_perm_t_sizes<0x7F, "vpermt2pd", defm VPERMT2PD : avx512_perm_t_sizes<0x7F, "vpermt2pd",
avx512vl_f64_info, avx512vl_i64_info>, VEX_W, EVEX_CD8<64, CD8VF>; avx512vl_f64_info, avx512vl_i64_info>, VEX_W, EVEX_CD8<64, CD8VF>;
//===----------------------------------------------------------------------===// //===----------------------------------------------------------------------===//
// AVX-512 - BLEND using mask // AVX-512 - BLEND using mask
// //
▲ Show 20 Lines • Show All 6,204 Lines • Show Last 20 Lines

test/MC/X86/avx512vbmi-encoding.s

// RUN: llvm-mc -triple x86_64-unknown-unknown -mcpu=skx --show-encoding %s | FileCheck %s
vpermi2b %xmm28, %xmm29, %xmm30
//CHECK: vpermi2b %xmm28, %xmm29, %xmm30
//CHECK: encoding: [0x62,0x02,0x15,0x00,0x75,0xf4]
vpermi2b %xmm28, %xmm29, %xmm30 {%k7}
//CHECK: vpermi2b %xmm28, %xmm29, %xmm30 {%k7}
//CHECK: encoding: [0x62,0x02,0x15,0x07,0x75,0xf4]
vpermi2b %xmm28, %xmm29, %xmm30 {%k7} {z}
//CHECK: vpermi2b %xmm28, %xmm29, %xmm30 {%k7} {z}
//CHECK: encoding: [0x62,0x02,0x15,0x87,0x75,0xf4]
vpermi2b (%rcx), %xmm29, %xmm30
//CHECK: vpermi2b (%rcx), %xmm29, %xmm30
//CHECK: encoding: [0x62,0x62,0x15,0x00,0x75,0x31]
vpermi2b 0x123(%rax,%r14,8), %xmm29, %xmm30
//CHECK: vpermi2b 291(%rax,%r14,8), %xmm29, %xmm30
//CHECK: encoding: [0x62,0x22,0x15,0x00,0x75,0xb4,0xf0,0x23,0x01,0x00,0x00]
vpermi2b 0x7f0(%rdx), %xmm29, %xmm30
//CHECK: vpermi2b 2032(%rdx), %xmm29, %xmm30
//CHECK: encoding: [0x62,0x62,0x15,0x00,0x75,0x72,0x7f]
vpermi2b 0x800(%rdx), %xmm29, %xmm30
//CHECK: vpermi2b 2048(%rdx), %xmm29, %xmm30
//CHECK: encoding: [0x62,0x62,0x15,0x00,0x75,0xb2,0x00,0x08,0x00,0x00]
vpermi2b -0x800(%rdx), %xmm29, %xmm30
//CHECK: vpermi2b -2048(%rdx), %xmm29, %xmm30
//CHECK: encoding: [0x62,0x62,0x15,0x00,0x75,0x72,0x80]
vpermi2b -0x810(%rdx), %xmm29, %xmm30
//CHECK: vpermi2b -2064(%rdx), %xmm29, %xmm30
//CHECK: encoding: [0x62,0x62,0x15,0x00,0x75,0xb2,0xf0,0xf7,0xff,0xff]
vpermi2b %ymm28, %ymm29, %ymm30
//CHECK: vpermi2b %ymm28, %ymm29, %ymm30
//CHECK: encoding: [0x62,0x02,0x15,0x20,0x75,0xf4]
vpermi2b %ymm28, %ymm29, %ymm30 {%k7}
//CHECK: vpermi2b %ymm28, %ymm29, %ymm30 {%k7}
//CHECK: encoding: [0x62,0x02,0x15,0x27,0x75,0xf4]
vpermi2b %ymm28, %ymm29, %ymm30 {%k7} {z}
//CHECK: vpermi2b %ymm28, %ymm29, %ymm30 {%k7} {z}
//CHECK: encoding: [0x62,0x02,0x15,0xa7,0x75,0xf4]
vpermi2b (%rcx), %ymm29, %ymm30
//CHECK: vpermi2b (%rcx), %ymm29, %ymm30
//CHECK: encoding: [0x62,0x62,0x15,0x20,0x75,0x31]
vpermi2b 0x123(%rax,%r14,8), %ymm29, %ymm30
//CHECK: vpermi2b 291(%rax,%r14,8), %ymm29, %ymm30
//CHECK: encoding: [0x62,0x22,0x15,0x20,0x75,0xb4,0xf0,0x23,0x01,0x00,0x00]
vpermi2b 0xfe0(%rdx), %ymm29, %ymm30
//CHECK: vpermi2b 4064(%rdx), %ymm29, %ymm30
//CHECK: encoding: [0x62,0x62,0x15,0x20,0x75,0x72,0x7f]
vpermi2b 0x1000(%rdx), %ymm29, %ymm30
//CHECK: vpermi2b 4096(%rdx), %ymm29, %ymm30
//CHECK: encoding: [0x62,0x62,0x15,0x20,0x75,0xb2,0x00,0x10,0x00,0x00]
vpermi2b -0x1000(%rdx), %ymm29, %ymm30
//CHECK: vpermi2b -4096(%rdx), %ymm29, %ymm30
//CHECK: encoding: [0x62,0x62,0x15,0x20,0x75,0x72,0x80]
vpermi2b -0x1020(%rdx), %ymm29, %ymm30
//CHECK: vpermi2b -4128(%rdx), %ymm29, %ymm30
//CHECK: encoding: [0x62,0x62,0x15,0x20,0x75,0xb2,0xe0,0xef,0xff,0xff]
vpermt2b %xmm28, %xmm29, %xmm30
//CHECK: vpermt2b %xmm28, %xmm29, %xmm30
//CHECK: encoding: [0x62,0x02,0x15,0x00,0x7d,0xf4]
vpermt2b %xmm28, %xmm29, %xmm30 {%k7}
//CHECK: vpermt2b %xmm28, %xmm29, %xmm30 {%k7}
//CHECK: encoding: [0x62,0x02,0x15,0x07,0x7d,0xf4]
vpermt2b %xmm28, %xmm29, %xmm30 {%k7} {z}
//CHECK: vpermt2b %xmm28, %xmm29, %xmm30 {%k7} {z}
//CHECK: encoding: [0x62,0x02,0x15,0x87,0x7d,0xf4]
vpermt2b (%rcx), %xmm29, %xmm30
//CHECK: vpermt2b (%rcx), %xmm29, %xmm30
//CHECK: encoding: [0x62,0x62,0x15,0x00,0x7d,0x31]
vpermt2b 0x123(%rax,%r14,8), %xmm29, %xmm30
//CHECK: vpermt2b 291(%rax,%r14,8), %xmm29, %xmm30
//CHECK: encoding: [0x62,0x22,0x15,0x00,0x7d,0xb4,0xf0,0x23,0x01,0x00,0x00]
vpermt2b 0x7f0(%rdx), %xmm29, %xmm30
//CHECK: vpermt2b 2032(%rdx), %xmm29, %xmm30
//CHECK: encoding: [0x62,0x62,0x15,0x00,0x7d,0x72,0x7f]
vpermt2b 0x800(%rdx), %xmm29, %xmm30
//CHECK: vpermt2b 2048(%rdx), %xmm29, %xmm30
//CHECK: encoding: [0x62,0x62,0x15,0x00,0x7d,0xb2,0x00,0x08,0x00,0x00]
vpermt2b -0x800(%rdx), %xmm29, %xmm30
//CHECK: vpermt2b -2048(%rdx), %xmm29, %xmm30
//CHECK: encoding: [0x62,0x62,0x15,0x00,0x7d,0x72,0x80]
vpermt2b -0x810(%rdx), %xmm29, %xmm30
//CHECK: vpermt2b -2064(%rdx), %xmm29, %xmm30
//CHECK: encoding: [0x62,0x62,0x15,0x00,0x7d,0xb2,0xf0,0xf7,0xff,0xff]
vpermt2b %ymm28, %ymm29, %ymm30
//CHECK: vpermt2b %ymm28, %ymm29, %ymm30
//CHECK: encoding: [0x62,0x02,0x15,0x20,0x7d,0xf4]
vpermt2b %ymm28, %ymm29, %ymm30 {%k7}
//CHECK: vpermt2b %ymm28, %ymm29, %ymm30 {%k7}
//CHECK: encoding: [0x62,0x02,0x15,0x27,0x7d,0xf4]
vpermt2b %ymm28, %ymm29, %ymm30 {%k7} {z}
//CHECK: vpermt2b %ymm28, %ymm29, %ymm30 {%k7} {z}
//CHECK: encoding: [0x62,0x02,0x15,0xa7,0x7d,0xf4]
vpermt2b (%rcx), %ymm29, %ymm30
//CHECK: vpermt2b (%rcx), %ymm29, %ymm30
//CHECK: encoding: [0x62,0x62,0x15,0x20,0x7d,0x31]
vpermt2b 0x123(%rax,%r14,8), %ymm29, %ymm30
//CHECK: vpermt2b 291(%rax,%r14,8), %ymm29, %ymm30
//CHECK: encoding: [0x62,0x22,0x15,0x20,0x7d,0xb4,0xf0,0x23,0x01,0x00,0x00]
vpermt2b 0xfe0(%rdx), %ymm29, %ymm30
//CHECK: vpermt2b 4064(%rdx), %ymm29, %ymm30
//CHECK: encoding: [0x62,0x62,0x15,0x20,0x7d,0x72,0x7f]
vpermt2b 0x1000(%rdx), %ymm29, %ymm30
//CHECK: vpermt2b 4096(%rdx), %ymm29, %ymm30
//CHECK: encoding: [0x62,0x62,0x15,0x20,0x7d,0xb2,0x00,0x10,0x00,0x00]
vpermt2b -0x1000(%rdx), %ymm29, %ymm30
//CHECK: vpermt2b -4096(%rdx), %ymm29, %ymm30
//CHECK: encoding: [0x62,0x62,0x15,0x20,0x7d,0x72,0x80]
vpermt2b -0x1020(%rdx), %ymm29, %ymm30
//CHECK: vpermt2b -4128(%rdx), %ymm29, %ymm30
//CHECK: encoding: [0x62,0x62,0x15,0x20,0x7d,0xb2,0xe0,0xef,0xff,0xff]